Largest AI Data Centers in the World
Ranked by disclosed power capacity (megawatts). Includes operational, under-construction, and planned facilities with confirmed public capacity figures. Many hyperscale campuses have not disclosed final capacity — this list covers only facilities with known figures. Download full dataset →
Showing facilities with disclosed power capacity figures only. Many facilities omit specific megawatt data. Browse the full directory →
About this ranking
Power capacity — measured in megawatts (MW) or gigawatts (GW) — is the primary metric for comparing AI data center scale. Unlike traditional data centers sized for general compute, AI facilities are constrained by available electricity rather than floor space. A single NVIDIA H100 GPU draws roughly 700 watts; a rack of 8 draws ~5.6 kW; a cluster of 100,000 GPUs demands roughly 56–100 MW depending on utilization and cooling overhead.
The facilities in this list represent a step-change in infrastructure scale. Meta's Hyperion campus in Louisiana, the Stargate joint venture, and Microsoft's Project Fairwater in Wisconsin are all designed to exceed the power consumption of major cities. The global AI buildout is driving unprecedented demand on power grids, pushing operators toward nuclear energy (SMRs, recommissioned plants) and large-scale renewable PPAs to secure reliable capacity.
Capacity figures reflect publicly disclosed targets — either from press releases, regulatory filings, or investigative reporting. Where a facility is under construction or planned, the figure represents the announced design capacity at full buildout, not current operational load.